In this work, way to model a production process from the process measurements is presented. The artificial neural network, with which the process model is built, is the Self-Organizing Map (SOM). The model is based on the capability of the SOM to build non-linear regression models of the data.
The attributes of individual products were used as process measurements. These attributes were used in training the Self-Organizing Map.
The model's ability to describe the process depends also on the attributes' capability to describe the behavior of the process. In case there are missing model variables, that is, some essential process variables not a part of the model, the model cannot be expected to give good results.
These attributes used in modeling include the element concentrations of the incoming raw material, the process parameter settings during the production of a particular product and quality characteristics of the end product.
With the aid of the model one can predict quality parameters and study the leverage effects of the process parameter changes. A software tool is presented to facilitate this. The method is applicable if there is a lot of data from the process. These results serve best if only a little is known about the behavior of the process. In this way, the model serves the process specialist in the learning of the essential from large amounts of measurement data.
The quality of the model itself will decrease as the the amount of noise in the measurements increases. Before any modeling technique can produce meaningful results, the inputs, that is, the measurements from the process must be consistent. Making the inputs consistent should be the first phase in any process improvement scheme [6]. It must be remembered that no modeling technique can compensate the lack of good data.